Companies Need More Workers. Why Do They Reject Millions of Resumes?
Companies are desperate to hire, and yet some workers still can’t seem to find jobs. Here may be one reason why: The software that sorts through applicants deletes millions of people from consideration.
Employers today rely on increasing levels of automation to fill vacancies efficiently, deploying software to do everything from sourcing candidates and managing the application process to scheduling interviews and performing background checks. These systems do the job they are supposed to do. They also exclude more than 10 million workers from hiring discussions, according to a new Harvard Business School study released Saturday.
Job prospects get tripped up by everything from brief résumé gaps to ballooning job descriptions from employers that lessen the chance they will measure up. Lead Harvard researcher Joseph Fuller cited examples of hospitals scanning résumés of registered nurses for “computer programming” when what they need is someone who can enter patient data into a computer.
Given the extreme labor shortages popping up across the economy, this article pretty reasonably raises the question of whether automated resume screening will and should continue. I believe that it will, even in the face of such a tight job market, as it emerges naturally from the market structure of digital labor markets.
Even though employers and job seekers have a high-level shared goal - get the right person hired - the incentives are misaligned such that hiring is an adversarial process, where each party is attempting to counter tactics of the other. Automated resume screening is used for triage, a key tactic for employers. This dynamic is self-reinforcing, and there is no reason to believe that temporarily tight labor markets will change the basic incentive scheme here.
Disclaimer: while I work in an adjacent software field at ZipRecruiter, the opinions expressed here are mine and do not represent those of my employer.
Time and attention are scarce resources
Triage is a key tactic for hiring at scale. Triage is most commonly used in the medical field, where it means prioritizing treatment of those most likely to recover. It’s appropriate when human or medical resources are extremely limited, in order to maximize the impact per unit of time spent or care delivered. In the context of recruiting, it often means prioritizing those candidates most likely to get hired.
Assume, if you will, an arbitrarily large stack of resumes placed before a recruiter. The recruiter is asked to go through this pile of resumes and pick out the five best resumes. The best way to do this is really clear - go through all the resumes, assess their quality, and take the best ones. There are no shortcuts when you’re looking for the best of anything. However, it is rare to find a situation where only the best will do, and so more likely a recruiter might be asked to pick out five very good resumes. In such a situation, triage can be invaluable.
If an algorithm can be asked to go through this stack of resumes and strike out the bottom half, the average quality of the remaining resumes will improve a lot. In a world where you want to pick the best resume, average quality doesn’t matter - but in a world where you want to pick some good ones, it does. Particularly if you assume that the stack of resumes is too big to manually review, applying an algorithmic screen cuts in half the time to review, and the chance of finding a good resume per hour doubles. An algorithmic resume screener could be simple and effective: for a nursing job, drop all resumes without a valid nursing certificate. This is what it might look like, in an example where the quality rating indicates “what is the percentage of this resume being a good fit”:
So the key insight into the ubiquity of resume screening systems, and the often bizarre results, is that this tactic works even if the screening algorithm is really bad. Let’s say the same algorithm is keyed to something correlated but not really relevant to job fit such as, say, “no gaps in work history”. The rationale here is that completely unqualified candidates are likely to have big gaps in their work history. You might find yourself tossing out a bunch of completely unqualified resumes but also lose some of the best. However, it does not necessarily harm the performance on the goal we set: we still cut the time required to screen resumes and still improve the average yield of an hour spent screening. The resulting pool is not as high-quality as a better screener, but still works better than doing nothing:
Whether with a good screening process or a bad one, the recruiter can still save a ton of time and improve their productivity. While the good screening process is more effective at producing the highest quality, a bad rule still improves quality and saves the same amount of human labor. It may not even substantially harm the overall goal, which is maximizing the number of potential hires.
Solve for the equilibrium
So it is natural that the use of automated screening techniques persists amongst hiring companies. Everyone is well aware that the practice is not “optimal”, in the sense that it is not likely to reliably produce the highest quality hire from a given pool of candidates. However, looking at it this way is an error, because that is not the margin at which companies are attempting to optimize.
Companies drive hiring at scale by saving time. This is Micro 101: to maximize the output of a given process from fixed inputs, focus on improving efficiency of whichever resource is most scarce. Candidates are generally plentiful, and so the scarcest resource for larger companies is the time recruiters and hiring managers spend reviewing applications and especially interviewing candidates. Winnowing candidates - even very crudely - saves hiring manager time, recruiter time, and improves the total throughput and total number of hires.
Candidates can get around this by applying to more jobs, which is the crux of the whole situation. As Tyler Cowen likes to say, “solve for the equilibrium” [Marginal Revolution]. If screening techniques lower the expected probability of being hired per application, you can improve your odds by applying more widely. This is why candidates are a plentiful resource, because as screening becomes more common it is increasingly rational for candidates to apply to more jobs. This in turn increases the returns to using screening technology and…well, you can solve for the equilibrium at home.
The resulting dynamic is what’s known as a Red Queen’s Race [Wikipedia], where competitors have to keep running faster to stay in the same place. While the tight labor markets make some changes to employer incentives at the margin, it is not sufficient to unwind the adversarial dynamic pointing towards growing use of resume screening. For white collar jobs, in fact, it is likely to become even more central with the rise of remote jobs. All of a sudden any remote-eligible job has a huge new pool of potential applicants (and resumes to screen), and any job seeker has a huge new pool of potentially eligible jobs.
The tight labor market might push us towards cutting back on algorithmic screening, but the Red Queen’s Race created by technology is too powerful. For better or worse, I expect the trend towards algorithmic resume screening is here to stay.